The Chinese People's Liberation Army, the PLA uses DeepSeek AI in non-combat missions. The Deep Seek can operate as the opponent and data seek missions in war games or military training.
The AI can be a good opponent for generals and admirals to research and develop tactics. That they want to use.
To be effective generals and admirals those people need good opponents in their war games. AI requires lots of information that they can benefit from in those missions. The DeepSeek can also used to train other AIs for military missions.
The problem is that. Open military AIs can be banned. And if mission-critical AI operates directly with outside users.
Those users can infect it using malicious code. And if that AI does not work right. That is a catastrophe.
And there is the possibility. In real cases, malware can infect those mission-intensive systems. Or as an example, electronic counter measures, ECM, and denial of service, DOS attacks can deny those LLM's operations.
So, the solution is to use open applications as gloves. That keeps the military AI sterile against malicious code.
If mission-critical AI operates directly with outsider users. They can see things like its server gates. The server that runs the LLM is similarly vulnerable to other servers. Hackers and malware can make attacks against those servers.
Malicious software can infect that large language model, LLM.
The open civilian application makes one extra layer between outside users and the military AI.
The DeepSeek can collect databases from multiple sources. And when it is ready.
The system can transfer those databases to other AI. The thing is that. The AI can use multiple sources to develop or train itself.
The training is like collecting some kind of encyclopedia of things that the AI must react to. Without information the AI is helpless. It must recognize details and then compile a reaction database with things that it sees.
The AI must recognize the situation and then find the action that allows it to make counter movement for that thing.
Those data sources can be sensors like spy satellites and recon planes. The system can also use things like computer games to create counteraction for some maneuvers. The AI can transfer the maneuvers that it detects from those sensors to the computer game. And then, the opponent who might not know about that information source starts to play against those maneuvers.
The system can render those objects like tanks to things. That even Western players can accept them. That kind of network-based data-collecting and training systems can teach those military AIs at a level. That they will not otherwise reach. The large-scale information collection is the thing that makes the AI more intelligent and capable to give response multiple situations.
https://interestingengineering.com/military/china-pla-deepseek-ai-non-combat-operations
Image: Interesting Engineering
Ei kommentteja:
Lähetä kommentti